Token TelemetrySee exactly what your coding & autonomous agents cost, think, and do.
Local, read-only observability for Claude Code, Codex, Gemini CLI, Cursor, Copilot, Antigravity, Qwen CLI, OpenCode, Grok Build, Hermes Agent, and Vibe — one command, no signup, your logs never leave your machine.
“Why did that Codex run cost $4.20?”
$curl -fsSL https://raw.githubusercontent.com/VasiHemanth/tokentelemetry/main/install.sh | bashMIT licensed · runs offline · requires Node 18+, Python 3.9+
The only observability built for autonomous agents.
Hermes runs across CLI, Telegram, Discord, Slack, Feishu, DingTalk, cron, webhook — 38 source platforms in total. TokenTelemetry is the only tool that observes them as a single agent, with a dedicated dashboard that respects how Hermes actually works.
$curl -fsSL https://tokentelemetry.com/install.sh | bash
$hermes plugins install VasiHemanth/tokentelemetry-hermes-plugin$hermes dashboard
The plugin is a launcher, not the engine. It opens TokenTelemetry pages inside Hermes Dashboard — but only when TT itself is running. Skip step 1 if you already have TT on:3000.
What you'll see
One dashboard. Every agent.
Autonomous agent observability — a different class of agent, its own surface.
Dedicated /hermes dashboard: gateway health, scheduled-job status, source breakdown across 38 platforms (CLI / Telegram / Discord / Slack / Feishu / DingTalk / cron / webhook / …).
Per-API-call latency and cache-hit % parsed from agent.log — none of the other agents emit this.
Subagent delegation rendered inline: each delegate_task call expands to show the child's summary, tokens, duration, and tool trace.
Skills + memory pages: 90 loaded skills with platform conditions, MEMORY.md / USER.md with char-limit progress bars.
Cost anomaly detection: silent reasoning-token waste (MiMo thinking-mode) flagged automatically.
Hermes Dashboard plugin: one-command install adds a TokenTelemetry tab inside Hermes's own web UI (port 9119) — deep-link launcher to Overview, Skills, Memory, Analytics, Projects.
Supported
Eleven agents. Zero config.
TokenTelemetry reads logs your agents already write. No proxies, no wrappers, no SDK to register.
FAQ
Common questions
What is Token Telemetry?
+
Token Telemetry (also written TokenTelemetry, sometimes misspelled as 'token telementry' or 'tokentelementry') is a free, open-source, 100% local observability dashboard for AI coding agents like Claude Code, Codex, Gemini CLI, Cursor, and GitHub Copilot. It tracks tokens, cost, tool calls, and reasoning by reading the log files those agents already write — no SDK, no signup, no cloud.
How do I track Claude Code token usage?
+
Install TokenTelemetry, run Claude Code normally, and open http://localhost:3000. TokenTelemetry auto-detects Claude Code sessions from ~/.claude/ logs — no instrumentation, no SDK, no config.
How do I monitor Gemini CLI and Codex costs?
+
TokenTelemetry auto-reads logs from Gemini CLI, OpenAI Codex CLI, Cursor, GitHub Copilot, Qwen CLI, OpenCode, Vibe, Antigravity, and Grok Build (xAI). Token counts and dollar costs appear in the local dashboard automatically.
Is there a free tool to monitor AI coding agent token usage?
+
Yes — TokenTelemetry is free, open-source (MIT), and runs 100% locally. No account, no signup, no cloud.
Does TokenTelemetry send my data to the cloud?
+
No usage data, ever. The dashboard reads session log files from your local filesystem and serves a UI on localhost — your logs, sessions, tokens, and costs never leave your computer. The only outbound call is an optional update check that fetches the latest version and release notes from GitHub (no usage data sent); turn it off in Settings → Updates & privacy, or with TT_NO_UPDATE_CHECK=1.
How does TokenTelemetry compare to Langfuse or Helicone?
+
TokenTelemetry is purpose-built for AI coding agents and is zero-config — no SDK instrumentation. Langfuse and Helicone are general LLM-app observability platforms that require code changes and (typically) a cloud account.
Which agents does it support?
+
Ten coding agents (Claude Code, OpenAI Codex, Gemini CLI, Cursor, GitHub Copilot, Qwen CLI, OpenCode, Vibe, Antigravity, Grok Build) plus Hermes Agent — Nous Research's autonomous agent, which gets its own dedicated dashboard at /hermes with gateway health, scheduled-job monitoring, skills + memory observability, and 38 source platforms (CLI / Telegram / Discord / Feishu / DingTalk / cron / webhook / …).
Why does Hermes Agent get its own page?
+
Hermes is structurally different from coding agents — it runs across messaging platforms (Telegram / Discord / Slack / WhatsApp / Signal / Matrix / Feishu / DingTalk / WeChat), supports persistent skills and memory, delegates to subagents, and runs scheduled cron jobs. Forcing it into the same UI as Claude Code would hide most of what it does, so it gets a dedicated surface that respects its shape.
Can I use TokenTelemetry from inside Hermes Dashboard?
+
Yes — there's a Hermes Dashboard plugin that registers a 'TokenTelemetry' tab inside Hermes's web UI at port 9119. It's a thin launcher: deep-link cards open the relevant TokenTelemetry page (Hermes Overview, Skills, Memory, Analytics, Projects) in a new browser tab, so you don't have to remember a second port. Install with `./scripts/install-hermes-plugin.sh` from the TokenTelemetry repo, then run `hermes dashboard`.
100% local
Your logs never leave your machine. No usage tracking. No accounts. Only an optional update check pings GitHub.
MIT open source
Read every line. Fork it. Replace it with something better — up to you.
No signup
One command, browser opens. That is the entire onboarding.
